An Adaptive Graph Morlet Wavelet Transform for Railway Wayside Acoustic Detection

Monitoring condition of train axle bearings is significant for ensuring the safety of train operation. Wayside acoustic detection technology is a promising tool for early fault detection of train axle bearings. However, the strong background noise exists in collected acoustic signals, which normally masks the fault feature of axle bearings. In this work, a novel method called adaptive graph Morlet wavelet transform (AGMWT), is proposed for fault feature extraction in railway wayside acoustic detection. In AGMWT, acoustic signals are firstly transformed into horizontal visibility graphs. Next, the inner product operation is conducted to measure the similarity between the graph and daughter wavelets which are translated and scaled versions of the mother Morlet wavelet. An adaptive wavelet threshold and a shrinkage strategy are then proposed to shrink the graph Morlet wavelet coefficient, and finally the denoised signal can be obtained using inverse transform. To improve denoising performance, parameters of the mother Morlet wavelet are then optimised according to the Hilbert envelope spectrum fault feature ratio and the Hilbert envelope entropy. The effectiveness of the proposed method has been verified by conducting simulation, laboratory and field experiments. In addition, the denoising ability of AGMWT has advantages comparing other methods.